Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Serverless Containers for app modernization

Serverless Containers for app modernization

When you build an app, you typically need to choose agility with serverless or flexibility with containers but not both. But why does it have to be that way? Wouldn't it be nice to have the best of both worlds?

In this presentation, we explore the open source project Knative and its managed version Cloud Run. We see how these projects enable you to deploy and manage containers in a serverless way on wherever you want, on-prem or in the cloud.

Once we understand the building blocks of Knative and Cloud Run, we look at a modernization case study where an old ASP.NET monolith was decomposed into a set of loosely-coupled, event-driven, Cloud Run microservices. We go through the transformation journey, explore the decision points, and deep dive into the final architecture with a discussion of pros/cons.


Mete Atamel

January 30, 2020


  1. Serverless Containers with Knative & Cloud Run How can they

    help for app modernization? Mete Atamel Developer Advocate at Google @meteatamel speakerdeck.com/meteatamel
  2. Introduction

  3. Operational Model Programming Model No Infra Management Managed Security Pay

    only for usage Service-based Event-driven Stateless Serverless
  4. Containers Any language Any library Ecosystem around containers .js .rb

    .go .py .sh … 0 1 0 1 0 0 1 1 1
  5. Containers Flexibility Serverless Velocity

  6. Cloud Run Fully managed, deploy your workloads and don’t see

    the cluster. Cloud Run on Anthos Deploy into Anthos, run serverless side-by-side with your existing workloads. Knative Everywhere Use the same APIs and tooling anywhere you run Kubernetes with Knative. Serverless containers with Knative and Cloud Run
  7. Knative

  8. Confidential & Proprietary What is Knative? Kubernetes based open source

    building blocks for serverless github.com/knative
  9. Knative and Cloud Run Stack Serving Eventing Kubernetes Platform Products

    Components Cloud Run Cloud Run on Anthos Istio Gateway *No eventing * *
  10. Confidential & Proprietary Knative Serving What is it? Rapid deployment

    of serverless containers Automatic (0-n) scaling Configuration and revision management Traffic splitting between revisions Pluggable Connect to your own logging and monitoring platform, or use the built-in system Auto-scaler can be tuned or swapped out for custom code
  11. Confidential & Proprietary Knative Serving Knative Service High level abstraction

    for the application Configuration Current/desired state of an application Code & configuration separated (a la 12-factor) Revision Point in time snapshots for your code and configuration Route Maps traffic to revisions
  12. Confidential & Proprietary Knative Eventing What is it? For loosely

    coupled, event-driven services with on/off cluster event sources Bind declaratively event sources, triggers and services Scales from just few events to live streams Uses standard CloudEvents Event type Flow Event source Event type Event type Event consumer(s)
  13. Confidential & Proprietary Knative Eventing Namespace subscribe Trigger Service (Callable)

    Broker Trigger Service (Callable) subscribe filter= filter= ✓ ✓ ✓ Source Events Source Events ingress ingress publish
  14. Confidential & Proprietary Knative Event Sources Name Description Apache Camel

    Allows to use Apache Camel components for pushing events into Knative Apache Kafka Brings Apache Kafka messages into Knative AWS SQS Brings AWS Simple Queue Service messages into Knative Cron Job Uses an in-memory timer to produce events on the specified Cron schedule. GCP PubSub Brings GCP PubSub messages into Knative GitHub Brings GitHub organization/repository events into Knative GitLab Brings GitLab repository events into Knative. Google Cloud Scheduler Google Cloud Scheduler events in Knative when jobs are triggered Google Cloud Storage Brings Google Cloud Storage bucket/object events into Knative Kubernetes Brings Kubernetes cluster/infrastructure events into Knative https://github.com/knative/docs/tree/master/docs/eventing/sources
  15. Confidential & Proprietary Knative Events { "specversion": "0.2", "type": "com.github.pull.create",

    "source": "https://github.com/cloudevents/spec/pull/123", "id": "A234-1234-1234", "time": "2019-04-08T17:31:00Z", "datacontenttype": "application/json", "data": "{ GitHub Payload... }" } FTP GitHub GCS Broker FTP Receive Adapter GitHub Receive Adapter GCS Receive Adapter CloudEvent
  16. Confidential & Proprietary Cloud Storage Events to Vision API Cloud

    Storage Bucket Cloud Storage -> Cloud PubSub -> Knative -> Vision API Cloud PubSub Topic Knative Eventing Knative Service Cloud Vision API Labels 1 2 3 4 5 6
  17. Cloud Run

  18. Container to production in seconds Natively Serverless One experience, where

    you want it Cloud Run Bringing serverless to containers
  19. HTTPS Endpoint Public • Website • API endpoint Private •

    Internal services • Async tasks • Mobile backend • Webhook
  20. Container contract Listen on on port $PORT (default 8080)

    HTTP server must start < 4 min (timeout → 504) Request time < 15 min (default → 5 min) Stateless (in-memory file system, doesn’t persist) Computation only within request (No background activity)
  21. Container resources 1 vCPU per container instance (configurable to 2vCPU)

    256 MiB of memory up to a max of 2 GiB (configurable) 80 concurrent requests per container (configurable 1-80) 1000 max containers by default (configurable 1-1000) Access to a Metadata Server Sandboxed by gVisor
  22. Pay per use CPU / Memory / Requests 100ms

  23. Billable time Instance Billable Time Request 1 Start Request 1

    End Request 2 Start Request 2 End Instance Time Billable Non-billable
  24. Concurrency: up to 80 concurrent requests concurrency = 1 concurrency

    = 80
  25. Pub/Sub triggered internal services Cloud Run Cloud Pub/Sub Queue Queue

  26. Storage triggered internal services Cloud Run Cloud Pub/Sub Queue Queue

    Queue Cloud Storage
  27. Scheduled services Command Line Interface (CLI) User Interface (UI) Scheduler

    API Cloud Run Cloud Scheduler
  28. Services part of async tasks Cloud Tasks user_registration user_levelcompleted user_inapppurchase

    user_statechange Daily activity metrics service User profile service Payment processing service Game state service
  29. An app modernization story Monolith to microservices with .NET Core,

    containers and Cloud Run
  30. Stage 0: Prototype (Late 2015) Reader Transformer Web API ASP.NET

    (4.6) Windows app on IIS hosting +Worked, easy to understand and deploy, cheap -Too much coupling, bad DevEx (FTP to see logs!), no redundancy, no persistence, no resilience
  31. Stage 1: Lift & Shift (2016/2017) Deploy to Compute Engine

    on Google Cloud +Easy to move, redundancy, load-balancing, possiblity of autoscaling, better DevEx with Stackdriver logging, VM snapshots etc. -Expensive Compute Engine Compute Engine Cloud Load Balancer
  32. Stage 2: Containarization (Early 2019) Re-write in ASP.NET Core (2.2),

    containerize w/ Docker & deploy to App Engine Flex (Linux) +Windows license fees out, hassle-free autoscaling, revision management -Still VM based (expensive), slow deploys App Engine
  33. Stage 3: Serverless (Mid 2019) Update to ASP.NET Core (3.0)

    & deploy to Cloud Run + Serverless billing (no more VMs), much cheaper, awesome DevEx (fast deploys, integrated logging, etc.) - Cold starts Cloud Run
  34. Stage 4: Monolith to microservices (Early 2020) Cloud Scheduler Reader

    (Private) Cloud Pub/Sub Cloud Storage Converter (Private) Web (Public) Cloud Firestore
  35. Stage 4: Monolith to microservices (Early 2020) Update to ASP.NET

    Core (3.1) & decompose to 3 microservices, introduce persistence + Loosely coupled architecture, ability to update individual pieces (eg. Converter), ability to use different languages, better utilization, persistence - Many moving parts, more expensive
  36. Lessons Learned • Transformation does not have to be all

    or nothing • Even simple lift & shift can have huge benefits • Non-optimal solutions can be a stepping stone to more optimal solutions • Monolith decomposition is hard! Need a good reason beyond separation of concerns
  37. @meteatamel github.com/meteatamel/knative-tutorial knative.dev github.com/meteatamel/cloudrun-tutorial cloud.google.com/run github.com/meteatamel/amathus Thank you!